API · v1 preview

One call between the model and the clinician.

Send the AI's draft plus the patient context. CAIVI checks it against the record, guidelines and formulary, then returns a disposition, CAI/CRI scores and an audit ID — in under 200 ms.

EHR / FHIR  →  AI model  →  [ CAIVI intercept ]  →  Clinician
                                  │
                                  └→ audit log · CAI/CRI dashboard

Request

POST https://api.caivi.com/v1/intercept
Authorization: Bearer <CAIVI_KEY>
Content-Type: application/json

{
  "setting": "ward",
  "patient": {
    "fhir_bundle_ref": "Patient/8842",
    "allergies": ["sulfonamide"],
    "egfr": 28,
    "medications": ["metoprolol 25 mg BID"]
  },
  "model_output": "Start Bactrim DS 1 tab BID for UTI."
}

Response

200 OK  ·  142 ms

{
  "disposition": "HARD_INTERCEPT",
  "cai": 41.2,
  "cri": 88.6,
  "findings": [{
    "type": "contra",
    "merp": "G-I",
    "claim": "Bactrim DS",
    "evidence": "Documented sulfonamide allergy",
    "safe_alternative": "Nitrofurantoin 100 mg BID x5d (check eGFR)"
  }],
  "audit_id": "aud_7f3c91"
}

Dispositions

PASSNo issues. Output is delivered to the clinician unchanged.

WARNOutput delivered with inline annotations the clinician must acknowledge.

HARD_INTERCEPTOutput blocked. A safe alternative and reason are returned instead.

Integration options

REST

Call /v1/intercept from your app before showing AI output.

FHIR proxy

Point your SMART-on-FHIR app at CAIVI; it forwards and validates transparently.

Batch audit

Upload historical outputs to /v1/audit for CAI/CRI benchmarking.

Preview specification — endpoints are illustrative and not yet live.